Skip to content

Dataset · hausmer

Cleaned posts from 62 Russian and Ukrainian Telegram channels

hausmer/tg-channels-mixed-clean

Over 720,000 PII-clean posts spanning humor, prose, news, and tech.

What's inside

The corpus includes the full publication history of 62 public Telegram channels totaling 721,221 PII-cleaned posts. Languages include Russian, Ukrainian, Belarusian, and English. Topics cover humor and memes, original prose and fanfiction, book media, news (including two official police channels), local transport channels, and a telecom channel. Data is available as a single split or in separate configurations per channel.

What you can do with it

  • Fine-tuning (SFT or pretrain) Russian and Ukrainian language models on social media texts
  • Training LoRAs for post generation across genres from news to humor and prose
  • Training message topic or sentiment classifiers for messengers
Tasks
Text generation
Languages
Ukrainian, Russian, Belarusian, English
Size
1M–10M rows
Downloads this month
34.3K
Likes
1
Access
on request on Hugging Face

How to load it

from datasets import load_dataset

ds = load_dataset("hausmer/tg-channels-mixed-clean")